ABSTRACT. As the frontier of machine learning applications moves further into human interaction, multiple concerns arise regarding automated decision-making. Two of the most critical issues are fairness and data privacy. On the one hand, one must guarantee that automated decisions are not biased against certain groups, especially those unprotected or marginalized. On the other hand, one must ensure that the use of personal information fully abides by privacy regulations and that user identities are kept safe. The balance between privacy, fairness, and predictive performance is complex. However, despite their potential societal impact, we still demonstrate a poor understanding of the dynamics between these optimization vectors. In this paper, we study this three-way tension and how the optimization of each vector impacts others, aiming to inform the future development of safe applications. In light of claims that predictive performance and fairness can be jointly optimized, we find this is only possible at the expense of data privacy. Overall, experimental results show that one of the vectors will be penalized regardless of which of the three we optimize. Nonetheless, we find promising avenues for future work in joint optimization solutions, where smaller trade-offs are observed between the three vectors.

ABSTRACT. Artificial intelligence (AI) has been widely used in many fields, from medical diagnosis tools to intelligent virtual assistants. However, there is no consensus on how to deal with ethical issues. Using a systematic literature review and an analysis of recent real-world news about AI-infused systems, we cluster existing and emerging AI ethics and responsibility issues into six groups - broken systems, hallucinations, intellectual property rights violations, privacy and regulation violations, enabling malicious actors and harmful actions, environmental and socioeconomic harms - discuss implications, and conclude that the problem needs to be reflected upon and addressed across five partially overlapping dimensions: Research, Education, Development, Operation, and Business Model. This reflection may be relevant to caution of potential dangers and frame further research at a time when products and services based on AI exhibit explosive growth. Moreover, exploring effective ways to involve users and civil society in discussions on the impact and role of AI systems could help increase trust and understanding of these technologies.

ABSTRACT. ML/AI is the field of computer science and computer engineering that arguably received most attention and funding over the last decade. Data is the key element of ML/AI, so it is becoming increasingly important to ensure that users are fully aware of the quality of the datasets that they use, and of the process generating them, so that possible negative impact on downstream effects can be tracked, analysed, and, where possible, mitigated. One of the tools that can be useful in this perspective is dataset documentation.

The aim of this work is to investigate the state of dataset documentation practices, measuring the completeness of the documentation of several popular datasets in ML/AI repositories. We created a dataset documentation schema—the Documentation Test Sheet (DTS)—that identifies the information that should always be attached to a dataset (to ensure proper dataset choice and an informed use), according to relevant studies in the literature. We verified 100 popular datasets from four different repositories with the DTS to investigate which information was present.

Overall, we observed a lack of relevant documentation, especially about the context of data collection and of data processing, highlighting a paucity of transparency.

ABSTRACT. Competition in Artificial Intelligence (AI) technologies is at its fiercest, pushing companies to move fast and cut corners regarding the risks to human rights and other societal impacts. Without simple methodologies and widely ac-cepted instruments is hard for an organization to adopt a safe pace on how to develop and deploy AI in a trustworthy way. This paper presents a Maturity Model for Responsible AI, inspired in the EU Ethics Guidelines for Trustworthy AI and another principles and codes of conduct. The core component is a self-assessment tool that generates a roadmap for organizations to improve their approach to AI related development, ena-bling a positive effect in their business value. It includes requirements to achieve Trustworthy AI, and the methods and the key practices that will enable the principles outlined. The result is a consistent and horizontal approach to all industries and functions, taking into consideration that a simple and generic Maturity Model specific for Responsible AI is still not available. The model presented in this paper is purposed to fill that gap. The model was first pre-tested in two organizations, then improved, then pre-tested again, and it will be applied in several other organizations. Its final version is planned to be published at the end of 2023.

ABSTRACT. The financial industry is a major player in the digital landscape and a key driver of digital transformation in the economy. In recent times, the financial sector has come under scrutiny due to emerging financial crises, particularly in high-risk areas like credit scoring models where standard AI models may not be fully reliable. This highlights the need for greater accountability and transparency in the use of digital technologies in Fintech. In this paper, we propose a novel approach to enhance the interpretability of AI models by knowledge augmentation using distillation methods. Our aim is to transfer the knowledge from black-box models to more transparent and interpretable models, e.g., decision-trees, enabling a deeper understanding of decision patterns. We apply our method to a credit score problem and demonstrate that it is feasible to use white-box techniques to gain insight into the decision patterns of black-box models. Our results show the potential for improving interpretability and transparency in AI decision-making processes in Fintech scenarios.

ABSTRACT. The combination of the phenomenon of overcrowding with inefficient management of resources is a major obstacle to the good performance of hospital units and consequent degradation of the medical service provided. With the intent of promoting a solution to combat the problem, this paper is made an analysis of the impact that the underutilization of resources has, more specifically of beds, verified through the present lack of planning of their allocation among the various specialties of the hospital, not respecting the needs of each one in question. For this, four different Optimization Techniques were analyzed to realize which presented better results in optimizing the allocation of beds in the Hospital units. Hill Climbing and the Genetic Algorithm stood out among the compared algorithms, the latter presenting greater consistency and a shorter computation time. When tested with real data from Centro Hospitalar do Tâmega e Sousa, for a particular date, attained a total of 0 wrongly allocated patients against 92. Such reduction promoted the efficiency of all management processes and allocation of beds in the different hospital specialties. This translates into better patient service, reduced waiting time, and staff workload, which means increased performance in all adjacent medical issues

ABSTRACT. Transthyretin (TTR)-related familial amyloid polyneuropathy (ATTRv) is a life-threatening autosomal dominant disease and the age of onset represents the moment when first symptoms are felt. Accurately predicting the age of onset for a given patient is relevant for risk assessment and treatment management. In this work, we evaluate the impact of combining prediction models obtained from neighboring time windows on prediction error. In this case, we propose two different averaging approaches and compare the results with single-algorithm runs, while bagging age-oriented models. Our results show that by aggregating predictions from neighbor models we decrease the average mean absolute error obtained by each base learner. Overall, the best results are achieved by regression-based ensemble tree models as base learners.

ABSTRACT. This work describes applying a transformer-based ranking solution to the specific problem of ordering ICD diagnoses codes. Taking advantage of the TFR-BERT framework and adapting it to the biomedical context using pre-trained and publicly available language representation models, namely BioBERT, BlueBERT and ClinicalBERT (Bio + Discharge Summary BERT Model), we demonstrate the effectiveness of such a framework and the strengths of using pre-trained models adapted to the biomedical domain. We showcase this by using a benchmark dataset in the healthcare field - MIMIC-III - showing how it was possible to learn how to sequence the main or primary diagnoses and the order in which the secondary diagnoses are presented. A window-based approach and a summary approach (using only the sentences with diagnoses) were also tested in an attempt to circumvent the maximum sequence length limitation of BERT-based models. BioBERT demonstrated superior performance in all approaches, achieving the best results in the summary approach.

ABSTRACT. AI applications in hospital frameworks can improve patient- care quality and efficient workflows and assist in digital transforma- tion. By designing Smart Hospital infrastructures, creating an efficient framework enables patient information exchange between hospitals, point of care, and remote patient monitoring. Deep learning (DL) solutions play important roles in these infrastructures’ digital transformation pro- cess and architectural design. Literature review shows that DL solutions based on Automatic Speech Recognition (ASR) and Natural Language Processing (NLP) are rising concerning clinical data digitalisation, pop- ulation health management, and improving patient care. Nevertheless, one of the literature’s shortcomings highlights the limited research us- ing these solutions in real-world medical environments. As part of smart hospitals, smart medical scribes have been presented in several studies as a promising solution. However, just a few studies have tested it in real settings. Moreover, it was limited to non-existent studies on non-English systems, even yet to be found similar studies for European Por- tuguese. The proposed study evaluates NLP-based solutions in real-life Portuguese clinical settings focused on patient care for Smart Healthcare applications.

ABSTRACT. In the milling process of micro-machining, the optimization process is one of the keys to reduce production cost. By monitoring the tool wear and detecting when it is no longer acceptable, the machining process can be adjusted more accurately. This research explores four approaches using different machine learning models to predict machining tool wear during the milling process. The study is based on a dataset created with a face milling operation on stainless steel (AISI 303) round material. The machining is divided into a number of stairs and is performed with a 3mm tungsten carbide. Three different types of sensors are used to measure the wearing process, with acoustic emission, accelerometers and axis currents. The better approach achieved a f1-score of 73% on five classes.

ABSTRACT. Real Estate Agents perform the tedious job of selecting and filtering pictures of houses manually on a daily basis, in order to choose the most suitable ones for their websites and provide a better description of the properties they are selling. However, this process consumes a lot of time, causing delays in the advertisement of homes and reception of proposals. In order to expedite and automate this task, Computer Vision solutions can be employed. Deep Learning, which is a subfield of Machine Learning, has been highly successful in solving image recognition problems, making it a promising solution for this particular context. Therefore, this paper proposes the application of Vision Transformers to indoor room classification. The study compares various image classification architectures, ranging from traditional Convolutional Neural Networks to the latest Vision Transformer architecture. Using a dataset based on well-known scene classification datasets, their performance is analyzed. The results demonstrate that Vision Transformers are one of the most effective architectures for indoor classification, with highly favorable outcomes in automating image recognition and selection in the Real Estate industry.

ABSTRACT. Machine learning has significant potential for optimizing var- ious industrial processes. However, data acquisition remains a major challenge as it is both time-consuming and costly. Synthetic data of- fers a promising solution to augment insufficient data sets and improve the robustness of machine learning models. In this paper, we investi- gate the feasibility of incorporating synthetic data into the training pro- cess of the injection molding process using an existing Long Short-Term Memory architecture. Our approach is to generate synthetic data by simulating production cycles and incorporating them into the training data set. Through iterative experimentation with different proportions of synthetic data, we attempt to find an optimal balance that maximizes the benefits of synthetic data while preserving the authenticity and rel- evance of real data. Our results suggest that the inclusion of synthetic data improves the model’s ability to handle different scenarios, with po- tential practical industrial applications to reduce manual labor, machine use, and material waste. This approach provides a valuable alternative for situations where extensive data collection and maintenance has been impractical or costly and thus could contribute to more efficient manu- facturing processes in the future.

ABSTRACT. In the last decades most efforts to catalog and characterize the built environment for multi-hazard risk assessment have focused on the exploration of census data, cadastral datasets, and local surveys. The first approach is only updated every 10 years and does not provide building locations, the second type of data is only available for restricted urban centers, and the third approach requires surveyors with an engineering background, which is cost-prohibitive for large-scale risk studies. It is thus clear that methods to characterize the built environment for large-scale risk analysis at the asset level are currently missing, which hampers the assessment of the impact of natural hazards for the purposes of risk management. Some recent efforts have demonstrated how deep learning algorithms can be trained to recognize specific architectural and structural features of buildings, which is needed for earthquake risk analysis. In this paper we describe how convolutional neural networks can be combined with data from OpenStreetMap and Google Street View to help develop exposure models for multi-hazard risk analysis. This project produced an original annotated dataset of approximately 5000 images of buildings from the parish of Alvalade (Lisbon, Portugal). The dataset was then used to train and test different Deep Learning networks for Building Exposure Models. The best results were obtained with ResNet50V2, InceptionV3 and DenseNet201, all with accuracies above 82\%. These results will support future steps on build and assessing Exposure Models for Risk Seismic Analysis.

ABSTRACT. Digital twin technology has gained considerable traction in recent years, with diverse applications spanning multiple sectors. However, due to the inherent complexity and substantial costs associated with constructing digital twins, systematic development methodologies are essential for fully capitalizing on their benefits. Therefore, this paper firstly provides an exhaustive synthesis of related literature, highlighting: (1) ten core advantages of implementing digital twin technology; (2) five primary domains in which digital twin applications have been prevalently employed; and (3) ten principal objectives of digital twin applications. Subsequently, we propose a seven-step digital twin application development process, encompassing: (i) Digital Twin Purposing; (ii) Digital Twin Scoping; (iii) Physical Twin Modeling; (iv) Calibration and Validation; (v) Application Logic Development; (vi) External System Integration; and (vii) Deployment and Operation. This structured approach aims to demystify the intrinsic complexity of twinned systems, ensuring that the deployment of digital twin-based solutions effectively addresses the target problem while maximizing the derived benefits.

ABSTRACT. With the growth of IoT (Internet of Things) technologies, there has been a significant increase in opportunities to enhance various aspects of our daily lives. One such application is the prediction of car park occupancy using car park movement data, which can be further improved by incorporating weather data. This paper focuses on investigating how weather conditions influence car park occupancy prediction and aim to identify the most effective prediction algorithm. To achieve more accurate results, the researchers explored two primary approaches: Classification and Regression. These approaches allow for a comprehensive analysis of the parking scenario, catering to both qualitative and quantitative aspects of predicting car park occupancy. In this study, a total of 24 prediction models, encompassing a wide range of algorithms, were induced. These models were designed to consider various details, including parking features, location specifics, time-related factors and, crucially, weather conditions. Overall, this study showcased the potential of leveraging IoT technologies, car park movement data, and weather information to predict car park occupancy effectively. By exploring both classification and regression approaches, each yielding accuracy and r2score values surpassing 85%.